I'm not sure it can. I've read many different descriptions and looked at the math, but it's a very different approach from the twin-based variance components estimation procedures I've managed to beat some understanding into my head of, and while I've worked with multilevel models & random effects in other contexts, the verbal descriptions of using multilevel models for estimating heritability just don't make sense to me. (Judging from Visscher's commentary paper, I may not be the only one having this problem.) I think my understanding of linear models and matrices may be too weak for it to click for me.